• 대한수학회지
• /
• 제59권2호
• /
• pp.217-233
• /
• 2022

Although machine learning shows state-of-the-art performance in a variety of fields, it is short a theoretical understanding of how machine learning works. Recently, theoretical approaches are actively being studied, and there are results for one of them, margin and its distribution. In this paper, especially we focused on the role of margin in the perturbations of inputs and parameters. We show a generalization bound for two cases, a linear model for binary classification and neural networks for multi-classification, when the inputs have normal distributed random noises. The additional generalization term caused by random noises is related to margin and exponentially inversely proportional to the noise level for binary classification. And in neural networks, the additional generalization term depends on (input dimension) × (norms of input and weights). For these results, we used the PAC-Bayesian framework. This paper is considering random noises and margin together, and it will be helpful to a better understanding of model sensitivity and the construction of robust generalization.

• 응용통계연구
• /
• 제17권1호
• /
• pp.75-88
• /
• 2004

서포트 벡터 기계(Support Vector Machines, SVMs) 알고리즘은 표본 점들과 분리 초평면 사이의 최소 거리를 최대화하는 것에 관심을 가져왔다. 본 논문은 모든 데이터 점들과 분리 초평면 사이의 거리들을 고려하는 total margin을 제안한다. 본 논문에서 제안하는 방법은 기존의 서포트 벡터 기계 알고리즘을 확장하고, 일반화 오차 경계를 개선하게 된다. 새롭게 제안하는 total margin알고리즘이 기존 방법들과의 비교를 통하여 더욱 우수한 수행능력을 가지고 있음을 수치 예제들을 통하여 확인할 수 있다.

• 충청수학회지
• /
• 제21권2호
• /
• pp.223-230
• /
• 2008

In this paper, we present an upper bound of the reciprocal sums of generalized subset-sum-distinct sequences with respect to the first terms of the sequences. And we show the suggested upper bound is best possible. This is a kind of generalization of [1] which contains similar result for classical subset-sum-distinct sequences.

• 충청수학회지
• /
• 제23권4호
• /
• pp.823-826
• /
• 2010

Let p be a prime. We generalized Sturm's bound to a group ${\Gamma}_{0}^{+}(p)$ of which the genus is zero.

• 충청수학회지
• /
• 제22권1호
• /
• pp.7-10
• /
• 2009

With the introduction of a new parameter $n{\geq}1$, Kim generalized an upper bound for the exponential function that implies the inequality between the arithmetic and geometric means. By a change of variable, this generalization is equivalent to exp $(\frac{n(x-1)}{n+x-1})\;\leq\;\frac{n-1+x^n}{n}$ for real ${n}\;{\geq}\;1$ and x > 0. In this paper, we show that this inequality is true for real x > 1 - n provided that n is an even integer.

• Korean Journal of Mathematics
• /
• 제12권2호
• /
• pp.117-125
• /
• 2004

We investigate some properties of functions of ${\Phi}{\Lambda}$-bound variation on a closed bounded interval, which is the generalization of bounded variation.

• 한국국방경영분석학회지
• /
• 제9권1호
• /
• pp.75-85
• /
• 1983

This study is concerned with the problem of routing vehicles stationed at a central depot to supply customers with known demands, in such a way as to minimize the total distance travelled. The problem is referred to as the vehicle routing problem and is a generalization of the multiple traveling salesmen problem that has many practical applications. A branch-and-bound algorithm for the exact solution of the vehicle routing problem is presented. The algorithm finds the optimal number of vehicles as well as the minimum distance routes. A numerical example is given.

• Journal of the Korean Data and Information Science Society
• /
• 제19권1호
• /
• pp.25-36
• /
• 2008

Error Correcting Output Codes (ECOC) can improve generalization performance when applied to multicategory classification problem. In this study we propose a new criterion to select hyperparameters included in ECOC scheme. Instead of margins of a data we propose to use the probability of misclassification error since it makes the criterion simple. Using this we obtain an upper bound of leave-one-out error of OVA(one vs all) method. Our experiments from real and synthetic data indicate that the bound leads to good estimates of parameters.

• 산업경영시스템학회지
• /
• 제13권22호
• /
• pp.65-69
• /
• 1990

Two different algorithms for traveling salesman problem(TSP) will be discussed. One is the engineering approach to the TSP. The other one is Branch-and-Bound algorithm to take advantage of the special structure of combinational problems. Also a generalization of TSP will be presented.

• 한국경영과학회:학술대회논문집
• /
• 한국경영과학회 1997년도 추계학술대회발표논문집; 홍익대학교, 서울; 1 Nov. 1997
• /
• pp.52-55
• /
• 1997

In this paper, both the generalization of one artificial variable technique to the general bound problem and the efficient implementation of the technique are suggested. When the steepest-edge method is used as a pricing rule in the simplex method, it is easy to update the reduced cost and the simplex multiplier every iteration. Therefore, one artificial variable technique is more efficient than Wolfe's method in which the reduced cost and simplex multiplier must be recalculated in every iteration. When implementing the one artificial variable technique on the LP problems with the general bound restraints on the variables, an arbitrary basic solution which satisfies the bound restraints is sought first, and the artificial column which adjusts the infeasibility is introduced. The phase one of the simplex method minimizes the one artificial variable. The efficient implementation technique includes the splitting, scaling, storage of the artificial column, and the cure of infeasibility problem.